DeepMind-Princeton paper shows LLMs causally use confidence to decide whether to answer
GoogleDeepMind · x · 2026-09-07
Researchers from Google DeepMind and Princeton (lead author Dharshan Kumaran) published an open-access Nature Machine Intelligence paper offering causal evidence that LLMs use confidence signals to drive behavior, not just passively report them.
Key points:
- A four-phase paradigm elicited baseline confidence without abstention, then probed abstention behavior
- LLMs apply an implicit threshold on internal confidence when abstaining; confidence effect sizes were roughly an order of magnitude larger than alternative mechanisms
- Causal evidence via activation steering: boosting confidence decreased abstention, suppressing it increased abstention, confirmed by mediation analysis
Conclusion: models genuinely use their own confidence to decide whether to answer or abstain, paralleling metacognition in biological systems.
More from Models
- "Abliterated models just have brain damage" — community pushes back on unaligned model hype — dhadfieldmenell · 2026-10-26
- dair-ai publishes primer on Cloudflare's Clairic System One extraction model — dair_ai · 2026-10-26
- MiniMax M3.1 Flash free window on 0G ends early, shifts to paid pricing — MiniMax_AI · 2026-10-26
- Unverified Grok Bot concept: connect banks and brokers as your personal CFO — nima_owji · 2026-10-26
- NVIDIA's Nemotron 3.5 Super VL scores just 1.94% on ARC-AGI-2 — teortaxesTex · 2026-10-26
- Fintech insiders' AI consensus: ChatGPT, Claude and Grokbot top the pack — LexSokolin · 2026-10-26